U.S. patent application number 13/912473 was filed with the patent office on 2013-12-12 for equalizing smoothing for a scan comparison to database.
The applicant listed for this patent is Matthew David Kelly, Thomas George Wright. Invention is credited to Matthew David Kelly, Thomas George Wright.
Application Number | 20130329975 13/912473 |
Document ID | / |
Family ID | 46605609 |
Filed Date | 2013-12-12 |
United States Patent
Application |
20130329975 |
Kind Code |
A1 |
Kelly; Matthew David ; et
al. |
December 12, 2013 |
EQUALIZING SMOOTHING FOR A SCAN COMPARISON TO DATABASE
Abstract
A method for facilitating comparison of image data from a given
medical imaging protocol with archive image data, includes
obtaining subject image data from a scan of a subject using the
medical imaging protocol, obtaining archive image data differing
from the subject image data in a given variable, determining a
modifier usable to reduce the difference between the archive and
subject image data, using the modifier to modify one of the subject
image data and the archive image data to reduce the difference, and
comparing the modified archive or subject image data with the
subject or archive image data.
Inventors: |
Kelly; Matthew David;
(Oxfordshire, GB) ; Wright; Thomas George;
(Oxfordshire, GB) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
Kelly; Matthew David
Wright; Thomas George |
Oxfordshire
Oxfordshire |
|
GB
GB |
|
|
Family ID: |
46605609 |
Appl. No.: |
13/912473 |
Filed: |
June 7, 2013 |
Current U.S.
Class: |
382/128 |
Current CPC
Class: |
G06T 2207/30016
20130101; G06T 2207/10104 20130101; G06T 7/0014 20130101 |
Class at
Publication: |
382/128 |
International
Class: |
G06T 7/00 20060101
G06T007/00 |
Foreign Application Data
Date |
Code |
Application Number |
Jun 8, 2012 |
GB |
1210124.2 |
Claims
1. A method for facilitating comparison of image data from a given
medical imaging protocol with archive image data, comprising:
obtaining subject image data from a scan of a subject using the
medical imaging protocol; obtaining archive image data differing
from the subject image data in a given variable; determining a
modifier usable to reduce the difference between the archive and
subject image data; using the modifier to modify one of the subject
image data and the archive image data to reduce the difference; and
comparing the modified archive or subject image data with the
subject or archive image data.
2. A method according to claim 1, comprising determining the
modifier by, for at least one of the archive and subject image
data: obtaining reference image data from a scan of a reference
object using the medical imaging protocol used to capture the
archive or subject image data; comparing the obtained reference
image data of the reference object to standard reference image data
for the reference object; and determining a factor modifying the
obtained reference image data to reduce a difference between the
obtained reference image data and the standard reference image
data.
3. A method according to claim 2, wherein determining the modifier
further comprise determining a modifying factor for the other of
the archive or subject image data, and combining the two modifying
factors to arrive at the modifier.
4. A method according to claim 1, wherein the modifier is a
Gaussian smoothing filter, and the factor is the full width at half
maximum of the Gaussian filter.
5. A method of establishing a series of archive image data objects
for comparison with current subject image data, comprising: (a)
obtaining a given archive image dataset of a previous subject; (b)
obtaining current image data from a scan of a reference object
using the medical imaging protocol used to capture the archive
image dataset; (c) comparing the current image data of the
reference object to the archive image data for the reference
object; and (d) determining a factor modifying the current image
data to reduce a difference between the current image data and the
archive image data.
6. A method according to claim 5, further comprising recording the
factor for the given archive image dataset.
7. A method according to claim 5, comprising repeating steps (a),
(b), (c) and (d) for each of a series of given archive image
datasets.
Description
BACKGROUND OF THE INVENTION
[0001] 1. Field of the Invention
[0002] The present invention is concerned with facilitating
comparison of a medical image to a database relating to content of
the medical image.
[0003] 2. Description of the Prior Art
[0004] Comparing medical images of a patient to a database of scans
acquired from normal subjects is a well-known technique for trying
to highlight abnormalities in the patient scan: Equivalently, the
patient could be compared to a database of scans showing a certain
pattern of disease to look for either similarities or differences
that can be used in the diagnostic process; these abnormalities can
then be used in the process of diagnosis.
[0005] The process of comparison generally takes the form of the
following steps: [0006] The patient's scan must be spatially
normalized, that is to say that it must be possible to form a
spatial correspondence between the patient's scan and the database
to which it is being compared [0007] The patient's scan must be
intensity normalized, that is to say that the intensity of voxels
in the patient's scan must be normalised to remove, for example,
differing rates of uptake in the tracer due to each patient's
unique metabolism.
[0008] Additionally, a degree of smoothing (e.g., Gaussian) may be
applied to the data to reduce noise and reduce some of the
differences in image appearance that arise from different scanners
and reconstruction algorithms, as well as from local physiological
variations between patients.
[0009] Once the patient's data is spatially normalized, intensity
normalized and smoothed, it can be compared to the database, where,
for example, the number of standard deviations each voxel data is
away from the mean can be computed: voxels whose data is
significantly different from the average may be an indication of
disease.
[0010] The database itself is constructed by following the same
process for each of the scans that will be included in it. Once all
scans have been spatially normalized, intensity normalized and
smoothed, statistics are computed for each voxel in the volume
across all the scans (e.g., mean and standard deviation), giving
rise to output data volumes containing these statistics, which
populate the database.
[0011] This type of analysis is most often performed in brain
scans, since the relative similarity of the anatomy of different
subjects makes the process of spatial and intensity normalization
simpler. However, the principle is not restricted to such an organ
once the process of normalizations can be resolved.
[0012] Different scanners (both different models from the same
manufacturer and models from different manufacturers) and different
reconstruction protocols (both algorithm type and parameterisation)
can result in significantly different degrees of smoothing in the
resulting image (i.e., reduction in resolution) that is finally
read by a clinician. When doing the kind of database comparison
analysis described above, matching the degree of smoothing across
the scans used to build the database, and also matching the degree
of smoothing in any new patient scan with that used in the database
construction is fundamental in trying to minimize diagnostically
irrelevant differences that arise simply from differences in
resolution between the particular scanner/reconstruction
combination used for the scan. For example, in scans from a
scanner/reconstruction combination that produces higher resolution
images, when compared to those used to build the database, small
regions of high uptake may be present that were not visible in the
database images purely due to the lower resolution.
[0013] A previously considered method is to use a fairly large,
constant, degree of smoothing, since adding additional smoothing is
less detrimental to the overall process than not using enough.
However, this assumption breaks down, for example, when the input
data is too dissimilar to the database, and in this situation a new
database must be generated using data that more closely matches the
new input data. Even when this is done, the fact that a constant
amount of smoothing is used means that each scan is not optimised
for comparison to the database, meaning that differences may not
show up as clearly as they otherwise might do.
[0014] The ability to reduce voxel-wise variance between brain PET
images from different scanners has recently been reported by Joshi
et al (2009) as part of a multi-centre observational study: the
Alzheimer's Disease Neuroimaging Initiative (ADNI). In this study,
the physical Hoffman brain phantom was imaged on a range of
different scanners, and the resulting images were further smoothed
with a Gaussian filter to smooth all images to a common resolution.
The additional smoothing factors were then applied to clinical data
and demonstrated a reduction in differences between the scans.
SUMMARY OF THE INVENTION
[0015] The present invention addresses the problem of how to choose
an appropriate amount of smoothing for each patient scan, such that
the smoothed scan looks "similar", and is of comparable resolution,
to the smoothed scans used in the construction of the database. The
same technique can also be used to ensure that all scans used to
build the database look "similar" enough.
[0016] The benefit of such embodiments is that the database can be
made more accurate by incorporating refined smoothing levels,
leading to potentially higher sensitivity of the database
comparison scheme. Moreover, the system can make easier the
inclusion of data that were not reconstructed using the same
protocol or acquired on the same scanner, therefore allowing more
data to be included in a database without having to impose too
stringent conditions on the reconstruction protocol or the scanner
on which it was acquired.
DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0017] Embodiments of this invention make use of some technology
defined in GB 2469569 or US2010/0290680, both of which are hereby
incorporated by reference herein, whereby an appropriate amount of
additional Gaussian filtering is determined for many scanner and
reconstruction protocol (algorithm and parameterisation)
combinations. The goal of the technology is to apply just the right
amount of smoothing to make the scans comparable for quantification
from regions of interest. However, once the Gaussian filter sizes
have been identified, they can be directly used to apply the
correct amount of smoothing to equalise the smoothness of the
scans, both within and in comparison to the database, as described
above.
[0018] For example, consider a database that has been constructed
using images from scanner/reconstruction combination "A" that have
then been further smoothed with a 10 mm FWHM Gaussian. A clinician
wants to compare a new dataset acquired with scanner/reconstruction
combination "B". If the filter required to adjust
scanner/reconstruction combination "A" to a given standard (as
described in GB 2469569) is 8 mm FWHM Gaussian, and the filter
required to adjust scanner/reconstruction combination "B" to the
same standard is 4 mm FWHM, the additional smoothing that should be
applied to the image being compared=
(10.sup.2+8.sup.2-4.sup.2)=12.2 mm FWHM.
[0019] The technology defined in GB 2469569 or US2010/0290680 uses
acquisitions of the NEMA IQ phantom to identify the appropriate
filter to compensate for differences in scanner/reconstruction
configuration. However, an alternative phantom, such as the Hoffman
brain, may be used if found to be more appropriate for this
application (e.g., neuroimaging).
[0020] As an alternative to producing a fixed normal database by
smoothing all images to a common resolution, the database could be
smoothed, with mean and standard deviation values computed on the
fly. The amount of smoothing could be determined based on the
dataset loaded for comparison. For example, a lower degree of
smoothing could be selected when comparing data from a high
resolution scanner, whereas a higher degree of smoothing could be
used for data from an older scanner. This would enable the
clinician to benefit from the improved resolution in their current
scanner.
[0021] Embodiments of the invention may include features such as
the following: [0022] A method for building a database from
captured medical image data of a number of subjects, including:
pre-processing scans of an appropriate test image acquired on a
comparable scanner model and reconstructed with comparable
reconstruction algorithm and parameters to obtain the appropriate
degree of smoothing; spatially normalizing each separate scan;
intensity normalizing each separate scan; applying the previously
computed degree of smoothing to each of the scans in turn;
computing statistics from the collection of normalized, smoothed
datasets. [0023] The pre-processing step uses an image of a NEMA
image quality phantom. [0024] The statistics computed are
voxel-wise mean and standard deviation. [0025] The statistics
computed are region-of-interest mean and standard deviation values.
[0026] A method for comparing a captured medical image data of a
subject to a previously computed database, including:
pre-processing a scan of an appropriate test image acquired on a
comparable scanner model and reconstructed with comparable
reconstruction algorithm and parameters to obtain the appropriate
degree of smoothing; spatially normalizing the subject scan;
intensity normalizing the subject scan; applying the previously
computed degree of smoothing to the subject scan; comparing the
normalized, smoothed dataset to the database. [0027] The number of
standard deviations away from the mean at each voxel is computed.
[0028] The number of standard deviations away from the mean of each
region of interest is computed.
[0029] Although modifications and changes may be suggested by those
skilled in the art, it is the intention of the inventor to embody
within the patent warranted hereon all changes and modifications as
reasonably and properly come within the scope of his contribution
to the art.
* * * * *